Regression Artifacts: Bronchodilator Response and Regression to the Mean

Abstract

In CHEST (June 2020), Cottee et al1Cottee A.M. Seccombe L.M. Thamrin C. King G.G. Peters M.J. Farah C.S. Bronchodilator response assessed by the forced oscillation technique identifies poor asthma control with greater sensitivity than spirometry.Chest. 2020; 157: 1435-1441Abstract Full Text Full Text PDF PubMed Scopus (15) Google Scholar conclude that bronchodilator response assessed by the forced oscillation technique identifies poor asthma control with greater sensitivity than spirometry. Methodologic issues with this paper need to be addressed. The asthma control score is positively correlated with FEV1,2Nathan R.A. Sorkness C.A. Kosinski M. et al.Development of the asthma control test: a survey for assessing asthma control.J Allergy Clin Immunol. 2004; 113: 59-65Abstract Full Text Full Text PDF PubMed Scopus (1787) Google Scholar and a low FEV1 means greater bronchodilator response as subsequently illustrated.3Eliasson O. Degraff Jr., A.C. The use of criteria for reversibility and obstruction to define patient groups for bronchodilator trials. Influence of clinical diagnosis, spirometric, and anthropometric variables.Am Rev Respir Dis. 1985; 132: 858-864PubMed Google Scholar The authors have rediscovered regression to the mean, a phenomenon known for > 130 years,4Campbell D.T. Kenny D.A. A Primer on Regression Artifacts. The Guilford Press, New York, NY1999Google Scholar and no conclusions about bronchodilator response or asthma control can be drawn from this paper. Regression to the mean, which affects measures of change, has been described as “an artifact that as easily fools statistical experts as lay people” and “as subtle as it is ubiquitous.”4Campbell D.T. Kenny D.A. A Primer on Regression Artifacts. The Guilford Press, New York, NY1999Google Scholar This artifact can be illustrated by creating two batches of numbers (A and B) with a random numbers generator (both A and B have equal numbers, the same mean and SD, and a normal distribution). Batch A is then subtracted from batch B, and the difference is always negatively correlated with batch A, with a correlation coefficient of approximately −0.7 (Fig 1). This occurs because low numbers in each batch are more likely to match up with a higher number in the other batch and vice versa; they regress to the mean. This is operative in FEV1 when the measure is repeated (after bronchodilator). The question: what is a treatment effect vs a statistical artifact? The error is compounded by calculating the response as percent of the initial value.3Eliasson O. Degraff Jr., A.C. The use of criteria for reversibility and obstruction to define patient groups for bronchodilator trials. Influence of clinical diagnosis, spirometric, and anthropometric variables.Am Rev Respir Dis. 1985; 132: 858-864PubMed Google Scholar Variable airways obstruction (going from asthma attack to normal) does not equate bronchodilator response when stable. Stable patients with asthma have near normal or normal FEV1 and do not improve with a bronchodilator. On the other hand, the fixed airways obstruction in COPD demonstrates a bronchodilator response.3Eliasson O. Degraff Jr., A.C. The use of criteria for reversibility and obstruction to define patient groups for bronchodilator trials. Influence of clinical diagnosis, spirometric, and anthropometric variables.Am Rev Respir Dis. 1985; 132: 858-864PubMed Google Scholar In concordance with this information, patients with normal spirometry have minimal to no bronchodilator response, showing nevertheless an decreasing gradient of response from the low to high FEV1 within that group.5Hegewald M.J. Townsend R.G. Abbott J.T. Crapo R.O. Bronchodilator response in patients with normal baseline spirometry.Respir Care. 2012; 57: 1564-1570Crossref PubMed Scopus (10) Google Scholar Allow me to suggest to CHEST and all other medical journals that in the future, all published papers containing before and after measurements be required to contain a paragraph stating that the results are subject to regression to the mean and an analysis of what constitutes a statistical artifact as opposed to real change. Bronchodilator Response Assessed by the Forced Oscillation Technique Identifies Poor Asthma Control With Greater Sensitivity Than SpirometryCHESTVol. 157Issue 6PreviewPersistent bronchodilator response (BDR) following diagnosis of asthma is an underrecognized treatable trait, associated with worse lung function and asthma control. The forced oscillation technique (FOT) measures respiratory system impedance, and BDR cutoffs have been proposed for healthy adults; however, the relevance in asthma is unknown. We compared BDR cutoffs, using FOT and spirometry, in asthma and the relationship with asthma control. Full-Text PDF ResponseCHESTVol. 158Issue 6PreviewWe thank Dr Eliasson for his interest in our article1 and raising awareness of an important research issue, namely the potential for statistical artifacts in clinical studies. Full-Text PDF